一键在 Manus 中运行任何 Skill

unblocked-tools-guide

星标10

分支1

更新时间2026年5月14日 23:23

Tool selection guide and legacy mapping for Unblocked search tools. Use this skill whenever you're unsure which Unblocked tool to call; when a tool call fails because a tool isn't found or isn't available; when you encounter references to deprecated tools such as data_retrieval, research_task, or unblocked_context_engine; or when a fine-grained tool like context_search_code, context_search_prs, context_query_issues, context_query_prs, or context_get_urls isn't available in your current environment. This skill explains what's available where and how to fall back gracefully.

安装

用 Codex 或 Claude 帮你安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。

在 Manus 中运行

来源

unblocked

unblocked/skills

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

相关职业SOC

基于 SOC 职业分类

软件开发工程师计算机与数学类职业·SOC 15-1252

SKILL.md

readonly

同仓库更多 Skills

同仓库

unblocked-context-get-urls

unblocked/skills

Direct URL content resolution via context_get_urls. Use this to fetch full content from one or more URLs the agent already has in hand — PRs, issues, docs, Slack messages, or arbitrary web pages that were surfaced by a prior search or provided by the user. TRIGGER when: you already have concrete URLs from a previous context_research / context_search_* result and want the full body, not just the title and URL; the user pasted one or more links and wants them summarized or analyzed; you need to resolve a Jira/Linear/PR/Slack link directly. DO NOT TRIGGER when: you need to search for content (use context_research or the context_search_* family); the URL is local (localhost, file://, private network) — those aren't reachable.

2026-04-2910

unblocked-context-query-issues

unblocked/skills

Structured, filtered issue retrieval via context_query_issues. Use this instead of context_search_issues when you need a precise filtered list — scoped by project and/or person — rather than semantic relevance matching. TRIGGER when: the user asks "what bugs are open in PROJECT", "what issues is Alice working on", "list in-progress tickets for the payments team", "issues I filed last month"; you need a definitive list rather than a ranked set of relevant items; you already know the project key or person's name and want to enumerate issues under that filter. DO NOT TRIGGER when: the question is conceptual ("is there an issue about X") or you don't have a project/person anchor — use context-search-issues for semantic matching instead.

2026-04-2910

unblocked-context-query-prs

unblocked/skills

Structured, filtered PR retrieval via context_query_prs. Use this instead of context_search_prs when you need a precise filtered list — scoped by project/repo and/or person — rather than semantic relevance matching. TRIGGER when: the user asks "what PRs did Alice open last week", "list merged PRs in payments-service this month", "PRs I'm waiting on review for", "what did the team ship yesterday"; you need a definitive list rather than a ranked set; you already know the repo name or person and want to enumerate PRs under that filter. DO NOT TRIGGER when: the question is conceptual ("why was this introduced", "what PR added feature X") — use context-search-prs for semantic matching instead.

2026-04-2910

unblocked-context-research

unblocked/skills

Unified context retrieval via context_research. Searches PRs, docs, messaging platforms, issue trackers, and code history to surface WHY code works the way it does, what was tried before, and what constraints apply. TRIGGER when: you need decision history, prior art, or conventions before modifying code; planning, scoping, or researching a topic; investigating a bug or incident; you need filtered lookups across issues, messages, or PRs (e.g., "open bugs in PROJECT", "what did X work on last week"); Grep/Glob/Read failed to find the referenced code — this skill searches across repos and systems beyond the current workspace. DO NOT TRIGGER when: you only need current implementation (not history or reasoning) — use Grep/Glob/Read directly.

2026-04-2910

unblocked-context-search-code

unblocked/skills

Code-only search across the current repo and connected repos via context_search_code. Use this instead of context_research when you want code results without Slack, Jira, or doc noise mixed in. TRIGGER when: Grep/Glob/Read can't find a referenced class, function, or module (it may live in another connected repo); you need semantic code search rather than a literal pattern match; you want to find how something is implemented or used across repos; the user asks to "find", "locate", or "search for" code. DO NOT TRIGGER when: local Grep/Glob/Read can find what you need; you also need PR history, team decisions, Slack context, or issue tracker data — use context_research instead.

2026-04-2910

unblocked-context-search-documentation

unblocked/skills

Documentation-only search via context_search_documentation. Use this instead of context_research when you want docs — wikis, READMEs, runbooks, API references, architecture docs, ADRs, and onboarding guides — without Slack, PR, or issue noise mixed in. TRIGGER when: looking for setup or configuration instructions; finding a runbook or operational procedure; understanding an API contract or event schema; locating an architecture decision record or design doc; the user asks "is there docs for X", "how do I set up Y", "is this documented", or "find the runbook for Z". DO NOT TRIGGER when: you need the current code state — use Grep/Glob/Read; you need Slack threads, PRs, or issues alongside docs — use context_research instead.

2026-04-2910

name

unblocked-tools-guide

description

Unblocked Tools Guide

Critical — Do Not Confuse MCP Surface With CLI Availability

The MCP server exposes only context_research and context_get_urls in most environments. The fine-grained tools (context_search_code, context_search_prs, context_search_issues, context_search_documentation, context_search_messages, context_query_prs, context_query_issues) are CLI-only — they do not appear in your MCP tool list even when they are fully available on the machine.

Therefore: not seeing context_search_* or context_query_* in your deferred/MCP tool list tells you nothing about whether they are available. You must check the CLI separately with command -v unblocked.

Skipping this check and declaring the tools "unavailable" based on the MCP surface is the single most common failure mode for this skill family.

Access Policy — CLI First, Then MCP, Then Stop

For every Unblocked call, follow this order:

Prefer the Unblocked CLI. Check availability once per session with command -v unblocked (or unblocked --help). Cache the result — do not re-probe on every call. If present, invoke the matching CLI subcommand directly.
Fall back to MCP only if the CLI is confirmed unavailable or a CLI call fails. Use the equivalent MCP tool (context_research, context_get_urls, etc.). On MCP-only, fine-grained tools are not exposed — fall back to context_research with a steering instruction (see routing table below).
If neither is available, stop and notify the user. Do not substitute with unrelated tools (web search, Grep-only guessing, etc.). Tell the user:

Unblocked is not available in this environment. See the setup docs at https://docs.getunblocked.com/unblocked-mcp/mcp-overview to install the CLI or configure the Unblocked MCP server, then retry.

The CLI is preferred because it exposes the full set of fine-grained tools, handles auth locally, and is more robust than MCP in most environments.

CLI ↔ MCP Tool Mapping

Purpose	CLI subcommand	MCP tool	CLI?	MCP?
Broad search across all sources	`unblocked context-research`	`context_research`	yes	yes
Resolve URL content	`unblocked context-get-urls`	`context_get_urls`	yes	yes
Code-only search	`unblocked context-search-code`	`context_search_code`	yes	no
PR descriptions and review discussions	`unblocked context-search-prs`	`context_search_prs`	yes	no
Issue tracker results	`unblocked context-search-issues`	`context_search_issues`	yes	no
Documentation (wikis, runbooks, ADRs)	`unblocked context-search-documentation`	`context_search_documentation`	yes	no
Messaging (Slack, Teams)	`unblocked context-search-messages`	`context_search_messages`	yes	no
Structured issue queries	`unblocked context-query-issues`	—	yes	no
Structured PR queries	`unblocked context-query-prs`	—	yes	no

Common flags (CLI):

context-research and context-search-*: --query <text> (required), --instruction <text> (optional); context-research additionally accepts --effort low|medium|high.
context-query-*: --query <text> (required), --projects <name...> (optional, array), --user-name <name> (optional).
context-get-urls: --urls <url...> (required, array). No --query.

If you're on MCP only and a fine-grained tool call fails with "tool not found", that's expected — fall back to context_research with an instruction that steers it toward the source type you want (see below).

Routing and Fallbacks

Use the preferred tool when available via the CLI. If the CLI is present but a fine-grained subcommand errors, or you're on MCP-only and the fine-grained tool isn't exposed, fall back to context_research and steer it with the --instruction / instruction parameter:

What you need	Preferred tool	`context_research` fallback instruction
Full picture across all sources	`context_research`	—
Code (semantic, cross-repo)	`context_search_code`	`"Prefer code and implementation results; deprioritize docs, issues, and messages"`
PR history and decision reasoning	`context_search_prs`	`"Prefer PR descriptions and review discussions; deprioritize other sources"`
Issue tracker (bugs, tasks, epics)	`context_search_issues`	`"Prefer issue tracker results; deprioritize code and messages"`
Docs (wikis, runbooks, ADRs)	`context_search_documentation`	`"Prefer documentation, wikis, and runbooks; deprioritize code and messages"`
Team chat (Slack, Teams)	`context_search_messages`	`"Prefer Slack threads and team conversations; deprioritize code and docs"`
Filtered issue enumeration (by project/person)	`context_query_issues`	`"Prefer issue tracker results filtered by project and assignee; enumerate rather than rank"`
Filtered PR enumeration (by repo/person)	`context_query_prs`	`"Prefer PR results filtered by repository and author; enumerate rather than rank"`
Resolve a known URL to full content	`context_get_urls`	— (this tool is available on MCP; no fallback needed)
Current local code	Grep / Glob / Read	—

Search vs. Query — Which to Use

The tool family has two shapes for the same sources:

context_search_* — semantic relevance matching. Use when the question is conceptual ("is there an issue about X?", "what PR introduced Y?").
context_query_* — structured, filtered enumeration. Use when you have precise filters (project key, repo, person) and want a definitive list ("open bugs in PAY assigned to Alice", "merged PRs last week in payments-service").

If you don't have a concrete project/repo/person anchor, reach for context_search_*. If you do, context_query_* gives cleaner, more deterministic results.

For detailed guidance on any tool, see the corresponding skill: unblocked-context-research, unblocked-context-search-code, unblocked-context-search-prs, unblocked-context-search-issues, unblocked-context-search-documentation, unblocked-context-search-messages, unblocked-context-query-issues, unblocked-context-query-prs, or unblocked-context-get-urls.